基于数据滤波的带协方差重置的递推贝叶斯算法

针对传统最小二乘算法计算量大、在有色噪声干扰下估计有误差的问题，提出了一种基于滤波技术的带协方差重置的递推贝叶斯算法。该算法首先使用一个动态非线性滤波器对输入输出数据进行滤波，然后使用贝叶斯方法进行参数估计。同时，为了加快参数的收敛速度，在算法中加入了一种新型的协方差重置策略。计算量分析表明，当过程模型和噪声模型的阶数分别为6和4的时候，所提算法可以减少约62.35%的计算量。仿真结果显示，所提算法与传统最小二乘算法在采样数据长度为3000时的估计误差分别为0.771%和1.118%。因此，所提算法具有较高的计算效率，并且可以给出精度较高的参数估计值。

Data filter based recursive Bayesian identification algorithm with covariance resetting

Traditional least squares identification algorithm requires much computational cost and its estimates are biased when the noise is colored. To overcome these shortcomings, this paper proposed a filter based recursive Bayesian identification algorithm with covariance resetting. In this algorithm, we firstly filtered the input and output data by a dynamics nonlinear filter and then used recursive Bayesian algorithm to estimate parameters. We also integrated a modified covariance resetting method to the algorithm. Analysis revealed that the proposed algorithm can reduce the computational burden by 62.35% compared with recursive Bayesian algorithm. Simulations indicated that the estimation errors of the two algorithms were 0.771% and 1.118% respectively. So the proposed algorithm has higher efficiency and can generate estimates with higher accuracy.